ES2536213T3 - Inhaler - Google Patents

Abstract

An inhaler (1) comprising a capsule housing (2) for containing a medicament capsule (4), an air flow path (6) through which the air flows during an air flow event from at least an air inlet (8) to an outlet (10), the air flow path (6) passes through the capsule housing (2), the capsule housing (2) is defined by at least one wall ( 22) and configured such that when there is a capsule (4) located in the capsule housing (4) and sufficient air flows along the air flow path (6) through the capsule housing (2), the capsule (4) moves inside the capsule housing (2), the inhaler (1) is characterized in that the inhaler includes a first sensor (16), a processor (18) and a power supply (20) to feed the processor (18), the first sensor (16) is arranged in the inhaler (1) so that it can detect the movement of the capsule (4) within l capsule housing (2) and generates a first signal indicative of said movement, the processor (18) receives the first signal from the first sensor and uses said first signal to determine whether the first signal is indicative of the presence or absence of a capsule ( 4) in the capsule housing (2) during an air flow event and generates a capsule signal indicative of the same.

Description

5

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

E12711598

04-05-2015

DESCRIPTION

Inhaler

The present invention is related to an inhaler, specifically to a capsule based inhaler.

There are many known types of inhaler, for which a user can inhale to receive a medication contained therein. Some inhalers contain multiple doses of medication, which can be accessed sequentially by a user, while others are based on capsules and require a user to insert at least one capsule into the device for each administration. It can be difficult to accurately monitor the way a user uses the device when it is not directly monitored, as some users do not accurately track its use. This can lead to poor compliance with a therapeutic regimen that is not obvious to a prescriber, so the cause of a persistent symptom is unclear. Dispensers have been proposed to allow a user or third parties to review the number of doses taken by a user, but this does not necessarily correspond precisely to the number of correct doses actually taken by the user. For example, the capsules may be removed from a dispenser, but never put into the inhaler, or put into the inhaler, but the medication may not be administered for any reason.

WO2005113042 describes a dry powder inhaler that dispenses medication from a capsule and WO2009075794 describes a different capsule based inhaler. WO2007101438 describes an inhaler that includes a flow sensor adapted to feel sound / vibration in the 1003000 Hz range from which a flow parameter is derived in a flow path of the inhaler. European Patent EP1898977 describes a dose counter device for an inhaler, which will record a dose administered from a canister in the inhaler, and at the same time substantially reduce the risk of falsely registering an unmanaged dose. GB2398065 describes a dispenser that includes a pressure sensor and processing means. The pressure sensor is capable of detecting a pressure signature produced in the dispensing of the medicine from a dispensing container and generating signals that are processed by the processor to analyze and compare said signals with one or more data sets containing indicative data of one or more control pressure signatures. The processing means are programmed to use a result of said comparison to detect the amount of medication dispensed in comparison to an intended volume of the measured dose volume.

The present invention provides an inhaler comprising a capsule housing for containing a medicament capsule, an air flow path through which air flows during an air flow event from at least one air inlet to an outlet, the air flow path passes through the capsule housing, a first sensor, a processor and a power source to power the processor, the capsule housing is defined at least by a wall and configured such that when there is a capsule located in the capsule housing and sufficient air flows along the air flow path through the capsule housing, the capsule moves into the capsule housing, the first sensor is disposed in the inhaler so that it can detect the movement of the capsule inside the capsule housing and generates a first signal indicative of said movement, the processor receives the first signal of the sensor and uses said first signal to determine if the first signal is indicative of the presence or absence of a capsule in the capsule housing during an air flow event and generates a capsule signal indicative of the same.

The inhaler is intended to allow the administration of medication from the capsule to an airway, for example the lung, of a patient. The medicament may be a dry powder, a liquid or other suitable formulation and may include one or more active components to treat one or more disease states. The medicament may include one or more non-active components that may be to stabilize, thicken or otherwise change one or more characteristics of the formulation. The medication may not include any active components, for example the medication may be a placebo.

The air flow path includes an inlet to allow air to enter the air flow path. The term "air" should be understood as including any suitable gas, for example a gas that can be provided to a patient who may not have an identical composition to air, for example oxygen enriched gas. The output of the air flow path may be a mouthpiece or nose piece through which the user inhales in order to receive the medication from the capsule.

An air flow event is when the air flows through the air flow path. This can be caused by a user who inhales through the inhaler, inhales for example through a mouthpiece, or nose piece, or it could be caused by a source of pressure that causes air, or other gas, to flow through of the air flow path from the entrance to the exit and inside the user. Typically the air flow through the inhaler will be in a range between 15 and 150 liters per minute.

The processor may be electronic, for example it may include one or more analog or digital integrated circuits, discrete circuits or programmable digital processors. The processor may require a power source, for example a power source, to function. The sensor may be electronic and may also require a source of electrical power to operate or it may be a passive sensor.

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

E12711598

04-05-2015

The signals generated by the sensor and / or the processor can be electromagnetic and can be a time-varying signal, for example a waveform, or it can be an electronic activation / deactivation signal

or high / low or any other suitable form of signal.

The sensor can be any suitable type of sensor capable of generating a signal that can be processed to provide the determination of whether a capsule is present in the inhaler. For example, an optical sensor could be arranged to monitor the capsule housing and a signal from said sensor could be processed to determine if the signal is indicative of capsule movement in the capsule housing. It is expected that a basic and trial and error algorithm could produce a suitable way to process such a signal.

In one embodiment, the inhaler includes a first sensor that is an impact sensor and the first signal is an impact signal. The capsule housing is defined by at least one wall and is configured such that as a capsule moves inside the capsule housing the capsule repeatedly impacts the at least one wall. The impact sensor is disposed in the inhaler so that it can detect the impacts of the capsule on the capsule housing wall and generate an impact signal indicative of each impact.

The first sensor is disposed in the inhaler so that it can detect the movement of the capsule inside the capsule housing. It can detect the movement directly, for example an optical sensor that sees the movement of the capsule. In an alternative embodiment the sensor can detect the movement by indirectly feeling a parameter that can be analyzed to determine the presence or absence of a characteristic linked to the movement of the capsule, for example the impact of the capsule with a wall, or a variation in the air flow pattern when the capsule moves through the air inlets or outlets.

The advantage of an impact sensor over, for example, an optical sensor, is that it is not necessary to arrange any part of the impact sensor in the air flow path, which can simplify the construction of the air flow path and can facilitate a retroinstallation of such a sensor in an existing inhaler design. An optical sensor would need at least one window in the air flow path, through which it can 'see' in the capsule housing and generate a signal so that the inhaler can process that signal to detect a capsule in the same. The impact sensor can be any suitable sensor, for example a pressure transducer, a microphone or a piezoelement. In one embodiment the sensor is a microphone disposed in the inhaler in a location where the impacts of the capsule on the capsule housing wall can be 'heard' or 'felt' by the sensor. To produce an adequate impact signal, the output of more than one type of sensor can be combined. It should also be noted that the capsule could be modified to be more easily detectable by a sensor, for example the capsule may include a metallic or magnetic part that could be detected by a suitable sensor. In another embodiment the sensor is a piezo element disposed in the inhaler in a location where the impacts of the capsule on the capsule housing wall can be 'heard' or 'felt' by the sensor.

The inhaler may also include a memory to store the capsule signal for one or more air flow events for later recovery. This could be any suitable form of memory and can be erasable

or permanent For example, the memory can be electronically readable and / or writable and / or rewritable and could include fast memory, RAM, EPROM. The memory can also record the first signal, the data about the moment the signal was generated and any other data. The inhaler may include additional sensors that could provide usage data of a prescriber to the user and such data could be stored in a memory for later recovery. Data could be associated with particular air flow events.

The inhaler may also include an output from which an external device, such as a computer, can access the capsule signal and / or the contents of a memory. The output may include a plug into which a communication cable can be inserted. Additionally, or as an alternative, the output could include a signal generator to generate and transmit a wireless signal that can be received by an external receiver. The output can be a wireless transmitter, for example a WiFi ™ transmitter.

The processor can analyze the first signal using one or more different algorithms. The processor can analyze the first sensor signal using a peak value finding algorithm and determine if the calculated peak frequency is within predetermined limits to produce a capsule signal. These limits will be determined on the basis of the typical rotating capsule frequency at the expected flow rates within the inhaler geometry. During an air flow event, it has been found that the frequency with which the capsule hits the wall of the capsule housing is substantially constant and therefore appropriate limits can be generated. To reduce the effects of signal noise in the detection of impact events, a peak value finding algorithm is used and the computational complexity is relatively low.

The processor can analyze the sensor's impact signal using a frequency domain discriminator algorithm and determine if the proportion of signal strength between two different predetermined frequency ranges is within the predetermined limits to produce a capsule signal. During an air flow event, it has been found that the impact signals differ at particular frequency intervals between signals with a capsule present and those that are not present. Comparison of the signal strength ratio between two different predetermined frequency ranges reduces the effects of signal noise.

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

55

E12711598

04-05-2015

The processor can analyze the sensor's impact signal using a two-variable statistical algorithm that calculates two statistical variables to characterize the signal and determine if the calculated statistical measurements are in a predetermined domain in a scatter plot of one variable versus the another in order to produce a capsule signal.

When performing a statistical analysis to calculate statistical variables to characterize the impact signal, it has been found that some measurements of the impact signal differ for signals with a capsule present and those that no capsule is present.

The kurtosis variable is a statistical variable potentially useful for this purpose. In a probability plot versus a particular variable (x), if the variable x is Gaussian, then K = 0. However, if K> 0 the tails of the distribution are fatter at the expense of the central peak. On the contrary, if K <0 then the distribution has thinner tails and a wider and flatter peak. K is thus a bidirectional measurement of absence of Gaussian character.

The kurtosis variable (K) can be used to detect capsule collision transients because these events tend to push the tails of the sample distribution outward in a visibly predictable manner that makes the result clearly non-Gaussian. The breathing noise is only very Gaussian. However, background noise when not inhaled through the inhaler has very low power (and therefore a relatively low variance (σ2)) and can have a very high kurtosis variable because even transient even very small ones can have a large proportional impact on the tails of the signals. This gives two types of signal that need to be distinguished:

 Breathing noise, capsule absent, (K low, σ2 low to medium)

 Breathing noise, capsule absent, (mean K, σ2 low to high)

The ratio between peaks and average value of the signal itself or of the square of the signal, can be used as a capsule absence signal that tends to have less high peaks (impact events) and therefore a lower ratio of peak to means, medium.

The calculated variables can be the kurtosis variable and the variance, or they can be the ratio between peaks and mean value of the square, or of the magnitude, of the impact signal and the variance.

For all these algorithms, the limits that can be used to classify signal types between 'capsule present' and 'capsule not present' will vary from one type of inhaler to another and can be determined using simple trial and error methods. It is likely that there are minimal variations for inhalers of the same type and therefore these limits can easily be calculated for one type of inhaler.

The capsule housing may have any suitable shape within which the capsule can be moved sufficiently to allow a sensor to produce the appropriate signals. The capsule housing may allow the capsule to move in any of the following ways, from there to there longitudinally, radially or rotationally, either by rotating completely or to a limited angular extent. The capsule housing may include a part that has a substantially cylindrical shape with a diameter longer than a capsule to be contained therein and a height greater than the diameter of the capsule, but less than the length of the capsule and The air flow path is arranged to cause the capsule to rotate inside the capsule housing. This arrangement allows the capsule to rotate around an axis that substantially passes through its diameter. The rotation can be in addition to a substantially random wobble created by the flow of air around other axes.

The inhaler may include at least one drive device that can be operated by a user to cause an opening element to open a capsule inside the inhaler. The inhaler may also include a drive sensor to feel the drive of the drive device and generate a drive signal. The processor can be arranged to receive the drive signal. The actuator can be a button coupled to an opening member, for example a piercing element or a cutting blade that is adapted to create an opening in the capsule to allow access to a medication contained therein. There may be two drive devices, each with an associated opening element so that in one capsule two openings can be created. The drive sensors can be pushbutton switches. Each drive device can be associated with a drive sensor, but this does not have to be the case. The drive sensors could be used to 'wake up' the rest of the electronics since pressing the buttons should be the action performed by a user only before inhaling through the device.

The processor may be arranged to generate a dose signal indicative of whether a user has followed a correct sequence of use for the inhaler. The processor can generate the dose signal based on the capsule signal and the drive signal, the order in which those signals were generated and the time between those signals.

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

E12711598

04-05-2015

In any of these examples, one or more filters can be applied to the signal of each sensor, before applying one or more of the algorithms. The filters may include any of a high pass filter, a low pass filter, a noise reduction filter or any other suitable filter.

The inhaler may be substantially similar or substantially the same as the capsule inhaler described in WO2005 / 113042.

It should be understood that in this specification and in the claims that follow, unless the context requires otherwise, the word "understand", or variations such as "comprises" or "comprising", implies the inclusion of the integer or stage indicated, or group of integers or stages.

The invention will now be described further, by way of example only, with reference to the following drawings, in which:

Figure 1 shows an inhaler;

Figure 2 shows a scatter plot of the kurtosis variable versus variance;

Figure 3 shows a scatter plot of the ratio between peaks and mean value versus variance;

Figure 4 shows a graph of an example of a frequency domain discriminant analysis;

Figures 5a and 5b show a graph showing an example of a peak retention analysis;

Figure 6 shows an example of electronic hardware based on digital processing; Y

Figure 7 shows an example of a signal processing algorithm.

Figure 1 shows an inhaler 1 comprising a capsule housing 2 containing a medicament capsule 4. The inhaler 1 comprises an air flow path 6 through which air flows during an air flow event. The air flow path 6 extends from at least one air inlet 8 to an outlet 10 and passes through the capsule housing 2. Entrance 8 enters capsule housing 2 away from a center line. In this example, an upper part 10 of the capsule housing 2 is substantially cylindrical and the air inlet 8 enters substantially tangentially to the capsule housing 2 to favor air swirls within the capsule housing 2. The upper part 10 of the capsule housing 2 has a substantially cylindrical shape with a diameter larger than a capsule 4 contained therein and a height greater than the diameter of the capsule, but less than the length of the capsule 4. The housing 2 of the capsule includes a bottom 12, or chest, on which the capsule 4 initially rests. The capsule 4 contains a dry powder drug formulation 14.

The inhaler 1 further comprises a sensor 16, in this case a microphone, located adjacent to the lower part 12 of the capsule housing 2. The sensor 16 is coupled to a processor 18 that is powered by a power source 20, in this case a battery.

The capsule housing 2 is defined by at least one wall 22 and is configured such that when there is a capsule 4 located in the capsule housing 2 and sufficient air flows along the air flow path 6, the capsule 4 is dragged to the top 10 of the capsule housing 2 and rotates in the air flow. When the capsule 4 is rotated, it repeatedly impacts the wall 22 and the sensor 16 is arranged so that it can detect these impacts within the capsule housing 2. The sensor 16 generates a signal indicative of the impacts. The processor 18 receives the signal from the sensor 16.

The inhaler 1 also includes a pair of actuation buttons 24 that are coupled to piercing members 26. The buttons 24 can be pressed by a user to cause the piercing members 26 to drill holes at the ends of a capsule 4 disposed in the bottom 12 of the capsule housing 2. There are drive sensors 28 that can generate drive signals indicative of whether the drive button 24 has been pressed or not.

The processor 18 receives the signals from the sensors 16, 28 and produces an output signal that may be indicative of the presence of a capsule during an air flow event, the actuation of the actuation buttons 24, the correct use of the inhaler (correct sequence and timing of the drive and a capsule that is present during an air flow event). The output of the processor 18 and / or the raw output of the sensors are stored in a memory 30 and can be accessed using an output 32, in this case a wireless transmitter.

It should be noted that with a microphone sensor a significant amount of noise can be detected in addition to the detection of the desired impact events. The noise may be ambient, or be caused by the flow of air through the inhaler. This noise can vary considerably in volume and type, so some way is needed to discriminate between an indicative signal of impact and one that does not indicate such impacts.

10

fifteen

twenty

25

30

35

40

Four. Five

E12711598

04-05-2015

To properly use the device, it is necessary for a user to load a capsule into the inhaler, press the buttons to pierce the capsule and then inhale through the device, so that the capsule is agitated and rotates in the air flow of such so that a powder medicine in it is dispensed from the capsule and dragged into the airflow to the patient.

The way in which the inhaler electronics could work is as follows:

one.

The user presses the buttons and the drive signals are received by the processor.

2.

The processor begins to sample data from the first sensor for a predefined period of time. The data is processed online according to one or more of the algorithms discussed in this memory. Intermediate data is stored.

3.

The plausibility of the intermediate data is checked. If necessary, the data from multiple approaches are compared.

Four.

The results are stored for later transmission.

Some examples of how the processor can process the impact signal are described below.

One way to detect an impact on the inhaler signal when it is filled with a capsule is to compare the signal with a particular threshold. For example, for low to medium flow rates, capsule impacts can be identified in the signal by the application of a threshold and assuming that each time a threshold is exceeded it has been caused by a capsule impact. If a sufficient number of impacts is found during signal processing, it can be determined that the signal is indicative of the presence of a capsule. The number of impacts depends on the frequency of rotation of the capsule that depends on the design of the inhaler and must be calibrated for each type of inhaler.

Another method to analyze the sensor signal is a statistical approach in which to characterize the signal, statistical variables are calculated. A capsule inside the inhaler creates a very clear impact chime with high signal peaks at a low frequency. This creates a characteristic distribution of amplitude in the signal.

For this analysis, the signal is first passed through a high pass filter (HPF) with Z transform in equation (1)

image 1

This has the double effect of (i) reducing low frequency noise and any DC deviation and (ii) increasing high frequency noise and capsule collision transients.

Then a sliding window algorithm is performed operating on N data samples (typically N = 2048), skipping N samples at a time to save. In each window the kurtosis variable K and the variance σ2 are calculated using equations (2) and (3). The sliding window that has the maximum power (empirically associated with maximum 'information' in the use cycle) generates the necessary output (K, σ2) of the detector. As it is assumed that the data has a zero average value after the HPF, the sums can be done instantly without the presence of the average value.

image2

image3

As described above, the kurtosis variable is thus a bidirectional measurement of the absence of Gaussian character. If a random variable x is Gaussian, then K = 0. However, if K> 0 the tails of the distribution are fatter at the expense of the central peak. On the contrary, if K <0 then the distribution has thinner tails and a wider and flatter peak. K

The kurtosis variable is suitable for detecting capsule impact events because these events tend to push the tails of the sample distribution out in a visibly predictable manner that makes the result clearly not Gaussian. The breathing noise is only much more Gaussian.

This gives two types of signal that need to be distinguished:

 Breathing noise, capsule absent, (K low, σ2 low to medium)

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

E12711598

04-05-2015

 Breathing noise, capsule absent, (mean K, σ2 low to high)

Simulation tests were performed in order to classify an input signal as belonging to one of two classes (H1 = breathing noise + capsule present, H0 = not H1). Some example results are shown in the scatter plot in Figure 2.

Shows all recorded data sets using an example of microphone type. In the scatter diagram two clear areas of results can be identified. Among them there is an area where there is no result. The two areas represent breathing noise with a rotating capsule and only breathing noise.

As mentioned, ambient noise has a lower kurtosis variable. When such a signal is added to the capsule + breathing noise signal, the total kurtosis variable becomes smaller. Therefore, the data points will move further down the scatter plot.

After calculating the variance and the kurtosis variable, a classification decision about the results must be made. This is done by checking in which of the three areas delineated by the lines in Figure 2 the data points fall.

What is above the upper line 'C' represents an indicative signal of a capsule present. What is below the bottom line 'D' represents only breathing noise. What is between the two lines represents a capsule with noise.

It should be noted that very high levels of ambient noise can mask capsule noise and therefore push data points from the capsule region to the capsule absence region. To detect such matters, the techniques described below can be used.

This algorithm is useful because it is robust, has low calculation complexity and has few memory requirements. It should be noted that the calculation of the kurtosis variable requires a fairly large dynamic range since the squares and squares of the squares must be calculated.

This technique works for sampling rates as low as a few kilohertz. Nyquist sampling is not required as long as the peaks can still be sampled.

The statistical method described above of higher order can be simplified. The objective of this method is to detect the presence of high peaks in the signal while the majority of the signal is quite low. This is done by calculating the kurtosis variable. A potentially simpler method is to use the ratio of peak to maximum value of the square of the signal.

For this method the signal is passed again through the high pass filter described in equation (1). Then, again the variance for windows of typically 2048 samples is calculated. Also, for each window the largest square of a signal sample is recorded. Its value is divided by the average value to become the ratio between peaks and average value.

The variance and the ratio between peaks and average value are then used as the variance and the previous kurtosis variable. This is illustrated in Figure 3.

The same processing is done as before, where the data points are classified according to their area in the scatter plot.

This method has the same few memory requirements as the higher order statistical method. It also requires fewer computations and has a smaller dynamic range. This simplifies operation in cheap and small low power processors that typically only offer fixed point computations.

Another algorithm that can be used to analyze the sensor signal is a frequency domain discriminant analysis. Figure 4 shows a comparison of the analysis of a signal of an air flow event (50 l / min) in the inhaler with a capsule present, Line A, and without it, Line B. It is evident that the frequency spectrum with Full and non-capsule capsule data have a similar amplitude in the frequency band from 1 kHz to 2.5 kHz, but very different at frequencies above 4 kHz due to clear signals of the capsule impacts within the inhaler.

This algorithm compares the signal energy in the band 1 -2.5 kHz with that above 3 kHz. This can be done by performing a fast Fourier transform and adding the energy in the different bands, or more simply in the time domain by using a combination of bandpass and highpass filters. These filters and the subsequent energy comparison can be implemented using analog or digital techniques. This algorithm is useful since there is a low computational complexity, if it is implemented in the time domain using filters.

From tests in several flows and with simulated noise and breathing profiles, it has been found that this method provides a reasonably robust method to detect the presence of a capsule in the inhaler,

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

E12711598

04-05-2015

even in the presence of high noise levels. It was observed that most of the spectral content of the ambient noise tested was below 1000 Hz and thus would not affect the calculation made here of energy ratio.

Another method is to apply a peak detection algorithm, which aims to identify all the peaks in the signal that were caused by the impacts of the capsule. Again, the signal is filtered with high pass as in equation (1). The rest of the algorithm can be performed on the square samples of the filtered signal or only on the filtered signal itself. The signal can be processed in a sliding window to allow the calculation of a profile over time, but this is not necessary.

For this algorithm, the amplitudes of a sample or its square are compared with a peak retention value. If the sample is greater than the peak retention value it is said that a new peak retention event has to occur. In this case the peak count is increased and the peak retention value is set to the sample value. However, if the following sample is smaller than the peak retention value, no peak retention event is observed, and the current peak retention value is reduced simply by multiplying it by an appropriate decay factor (in this event an appropriate value is approximately 0.99). The pseudocode for this algorithm is given below, in which d (k) is the k-th sample of data and pk_hold is the peak retention value.

IF D (K)> pk_hold

pk_hold = d (k)

peak_counter = peak_counter + 1

peak_event (k) = 1;

ELSE

pk_hold = 0.999 * pk_hold END

The pseudocode does not show that, in one embodiment of this algorithm, it is necessary to process at least 20 samples before it can be considered that the following peak retention event has occurred. This prevents a group of peak retention events from occurring around the beginning of a capsule impact event and ensures that each impact is counted only once.

Also, low level noise can be eliminated using a threshold. Only sample values above a threshold are considered valid peaks. This avoids counting many very small peaks that are not true capsule impacts.

Having found the peak retention events, the algorithm measures the time between each event, from which a fundamental frequency can be calculated. The number of occurrences of particular fundamental frequencies within 10 Hz bands is then counted. The results of tests with the current inhaler type show that due to well-defined capsule impact events, the impact signal of a capsule measurement full has more low frequency content than high frequency content, and thus comparing signal energy below 110 Hz with that above 300 Hz is a suitable measurement to differentiate between full and absence capsule events of capsule.

Figures 5a and 5b show the results of peak retention processing for a full capsule and no capsule measurement respectively. Both tests were performed with a flow of 20 l / min through the inhaler and the graphs show a signal amplitude on the vertical axis and the number of samples along the horizontal axis.

The symbol 'o' represents a peak retention event and the lines that join the 'o' show how they decay between each of these events. From the figures, it should be noted that prominent capsule impact peaks have been located for full capsule measurement, but for the capsule absence measurement the algorithm only places closely spaced peaks due to the nature of the shape. Noisy wave

Tests at higher flow rates (150 l / min) have shown that individual impact peaks are closer to each other, so there is more high frequency content in full capsule measurements and as such the ratio between energy per below 110 Hz and 300 to 1000 Hz is not as large as for measurements with lower flow rates, but is still usable.

While all the algorithms described so far provide good performance in ideal calm conditions, ambient noise or noise caused by inhaler handling can cause false results.

10

fifteen

twenty

25

30

35

40

Four. Five

fifty

E12711598

04-05-2015

In order to avoid false results, the following techniques can be used:

Driving noise can cause high individual peaks in the signal. These are quite similar to the spikes caused by capsule impacts. However, only a very limited number of peaks is caused p. ex. by the inhaler that falls or hits a hard surface. While statistical algorithms or the frequency domain discriminator cannot distinguish such events from capsule events, they can be supplemented by the peak retention method. A capsule classification present for a signal can only be considered valid if there is a sufficient number of peaks present in a time interval. Otherwise the result is classified as noise.

Strong background noises can disguise the signal peaks that are used to detect a capsule using the various algorithms. As a breath only lasts for a limited time, there is a time before and after the breath in which no capsule or breathing noise should be expected. Therefore, the first part of the signal after the buttons are pressed (typically 0.1 to 0.5 seconds after the button is pressed) and the last part before stopping evaluating the signal (typically after 10 to 30 seconds) can be used to check background ambient noise levels. If these are above a certain level that make the capsule detection algorithms unreliable, a noise result will be created.

Figure 6 shows the physical equipment used in the example. The signal from microphone 16 is passed to an analog high-pass filter 50 which is a simple first-order RC filter with a 3dB frequency of 1 kHz. From there the signal passes to an analog-to-digital converter (ADC) 52 that samples at 9.6 kHz and has a resolution of 12 bits. The ADC may be integrated in a microprocessor chip 54.

Figure 7 shows an example of the combination of algorithms that can be performed on the samples once they are sent to the microprocessor 54. First, the signal is divided into windows of length of 2048 samples in a window separation operation 56.

These are processed by a simple high pass filter 58. The simplest implementation is to subtract the previous sample from the current one. This eliminates DC deviations that might be present due to circuit issues in the ADC. In an example of a signal there may be an inhalation of approximately two seconds duration at the beginning of the signal followed by a period of silence. The search window has to be much longer than a breath since the time it takes the user between the capsule piercing and the inhalation is unknown.

Then the square 60 of the signal is made and 62 the average value of the square samples in all 2048 samples of the window is calculated. This is done in the upper branch of the algorithm diagram in Figure 7. Also, in the central branch 64 the highest square value is recorded. This can be done while the squares are computed or through a search on all square samples if they are stored in memory. After computing all the sample squares and their average value and finding their peak value, the ratio between peaks and average value 66 is calculated. The average value of squares (variance) and the ratio between peaks and average value are stored for this window. for a later classification.

The lowest branch in the algorithm diagram in Figure 7 counts the peaks within window 68. First, a threshold is used to eliminate small peaks caused by noise. The peak detection algorithm is then applied to find the peaks caused by capsule impacts.

This process is repeated for each window until all windows are processed. It is expected that there is a group of results in variance approximately zero and with a low ratio between peaks and average value. These are the results of windows that only include ambient noise, for example after inhalation. Windows during inhalation tend to produce results with greater variance and ratio between peaks and average value.

Ranking 70 begins with a search for the window with the highest variance. As this contains most of the signal energy, this gives the most reliable information in the presence of other noises. Other measures are possible such as the search for a continuous set of windows with the highest energy, as well as to achieve optimum reliability. For the window with the highest variance, the corresponding ratio between peaks and average value is sought in the results.

The result should now be classified by comparison with a set of thresholds. These thresholds were determined by performing many experiments with and without capsule for various flows. A scatter plot of these results of the experiments can typically be divided into four regions:

one.

The "capsule" is typically at the top. This is the region of the ratio between peaks and average value (PMR, in English peak-to-mean ratio) due to capsule peaks.

2.

"Silence". This region has a very low variance and a low PMR.

E12711598

04-05-2015

3.

"Without capsule." This region has a low variance and a low PMR. The maximum variance is much lower than that achieved for capsules since no capsule impact can increase the noise level.

Four.

"Capsule with noise". This region is between "Capsule" and "Absence of capsule". The results go

to this region if a capsule signal has been subjected to high levels of ambient noise. Since the ambient noise level 5 has a lower PMR than capsule noise, the total PMR is reduced.

Finally, two checks are made against two issues that may arise from the processing done so far. Noise from inhaler handling 72, p. ex. an accidental fall on a hard surface creates large spikes in the signal. These can dominate the variance and cause a very high PMR. This would lead to an erroneous classification as "Capsule". Such handling noise events typically show only 2 to 4 peaks.

10 per window, while a rotating capsule shows more than 10. Also, due to the duration of breathing, the capsule rotates for at least one second. Therefore, the number of peaks in five consecutive windows is added. If this sum is greater than 50, the "Capsule" classification is confirmed. Otherwise, the classification verdict is revised to "Noise".

In some cases, a low capsule signal can be masked by a loud ambient noise. Due to the

15 PMR low this would go to the "Capsule absence" region. To recognize this situation, 74 the variance is checked in the last processed window. If the silence threshold is larger than approximately twice, the classification verdict is changed to "Noise". This helps to ensure that false negative results are not reported or very few.

It should be understood that the invention has been described above by way of example only and that modifications can be made in detail without departing from the scope of the claims.

Claims (14)

5 10 fifteen twenty 25 30 35 40 Four. Five fifty 1. An inhaler (1) comprising a capsule housing (2) for containing a medicament capsule (4), an air flow path (6) through which the air flows during an air flow event from by at least one air inlet (8) at an outlet (10), the air flow path (6) passes through the capsule housing (2), the capsule housing (2) is defined by at least one wall (22) and configured in such a way that when there is a capsule (4) located in the capsule housing (4) and sufficient air flows along the air flow path (6) through the housing (2) of capsule, the capsule (4) moves inside the housing

(2)

 capsule, the inhaler (1) is characterized in that the inhaler includes a first sensor (16), a processor (18) and a power supply (20) to power the processor (18), the first sensor (16) is available in the inhaler

(one)

 so that it can detect the movement of the capsule (4) inside the capsule housing (2) and generates a first signal indicative of said movement, the processor (18) receives the first signal from the first sensor and uses said first signal to determine if the first signal is indicative of the presence or absence of a capsule (4) in the capsule housing (2) during an air flow event and generates a capsule signal indicative of the same.

2. An inhaler according to claim 1, wherein the first sensor (16) is an impact sensor and the first signal is an impact signal and the capsule housing is defined by at least one wall (22), the Capsule housing (2) is configured such that, when a capsule (4) moves inside the housing (2) capsule, the capsule (4) repeatedly impacts on the at least one wall (22), the impact sensor (16) is arranged in the inhaler (1) so that it can detect the impacts of the capsule on the capsule housing wall (22) and generate an impact signal indicative of each impact.

3.

An inhaler according to claim 1, wherein the inhaler further includes a memory (30) for storing the capsule signal for one or more air flow events for later recovery.

Four.

An inhaler according to claim 1 or claim 2, wherein the inhaler further includes an outlet

(32) from which the capsule signal or memory content can be accessed.

5.

An inhaler according to any preceding claim, wherein the processor (18) analyzes the impact signal of the sensor (16) using a peak detection algorithm and determines whether the calculated peak frequency is within predetermined limits in order to Produce a capsule signal.

6.

An inhaler according to any preceding claim, wherein the processor (18) analyzes the sensor impact signal (16) using a frequency domain discriminator algorithm and determines whether the proportion of signal strength between two different predetermined frequency ranges is within the predetermined limits to produce a capsule signal.

7.

An inhaler according to any preceding claim, wherein the processor (18) analyzes the sensor impact signal (16) using a two-variable statistical algorithm that calculates two statistical variables to characterize the signal and determines whether the calculated statistical measurements are found in a predetermined domain in a scatter plot of one variable against the other in order to produce a capsule signal.

8.

An inhaler according to claim 7, wherein the variables calculated are the kurtosis variable and the variance.

9.

An inhaler according to claim 7, wherein the calculated variables are the ratio between peaks and average value of the square or the magnitude of the impact signal and the variance.

10.

An inhaler according to any preceding claim, wherein the processor (18) analyzes the impact signal of the sensor (16) using at least two different algorithms.

eleven.

An inhaler according to any preceding claim, wherein the first sensor (16) is a microphone or a piezo element.

12.

An inhaler according to any preceding claim, wherein the capsule housing (2) includes a part having a substantially cylindrical shape with a diameter longer than a capsule (4) to be contained therein and a height greater than the diameter of the capsule (4), but less than the length of the capsule and the air flow path (6) is arranged to cause the capsule to rotate inside the capsule housing.

13.

An inhaler according to any preceding claim, wherein the inhaler includes at least one actuator (24) that can be operated by a user to cause an opening element

(26) open a capsule inside the inhaler, the inhaler also includes a drive sensor (28) to feel the drive of the drive device (24) and generate a drive signal, the processor is arranged to receive the drive signal. eleven 14. An inhaler according to claim 13, wherein the processor (18) is arranged to generate a dose signal indicative of whether a user has followed a correct sequence of use for the inhaler, the processor (18) generates the signal of dose based on the capsule signal and the drive signal, the order in which those signals were generated and the time between the signals. 12